Let random seed setting work on xpu device (#2270)

Summary:
This Pull Request relates to [Roadmap Issue https://github.com/pytorch/benchmark/issues/1293](https://github.com/pytorch/benchmark/issues/1293) by enhancing our benchmark coverage.

Currently, Torchbench utilizes a custom random seed function that is incompatible with the XPU device backend.
This incompatibility affects models that include random data augmentation operations, leading to accuracy check failures due to variations in input data across two separate runs.

In this PR, we introduce support for setting a random seed for the XPU backend.

Pull Request resolved: #2270

Reviewed By: davidberard98

Differential Revision: D57884498

Pulled By: xuzhao9

fbshipit-source-id: 24234674333945782b233191e42a8b344e90d74c

Author: retonym

torchbenchmark/util/env_check.py

@@ -172,9 +172,12 @@ def deterministic_torch_manual_seed(*args, **kwargs):
 
         seed = MAIN_RANDOM_SEED
         import torch.cuda
-
         if not torch.cuda._is_in_bad_fork():
             torch.cuda.manual_seed_all(seed)
+
+        import torch.xpu
+        if not torch.xpu._is_in_bad_fork():
+            torch.xpu.manual_seed_all(seed)
         return default_generator.manual_seed(seed)
 
     torch.manual_seed(MAIN_RANDOM_SEED)